...
During the testing CloudBolt (an interface from which people can order self service VMs) configuration was applied to our configuration management (Puppet) server, which changed configuration on some nodes pointing them to wrong environment (test instead of production). CloudBolt in this configuration, is seen by Puppet as ENC (External Node Classifier). The ENC pushes some configuration to all the servers managed by puppet, and when you use a ENC, you cannot switch the configuration back, and it must be done from CloudBolt. This caused the services to pick the configuration from wrong environment that caused the outage.
...
Incident severity: CRITICAL
Data loss: NO
Affected Services
Following services were inaccessible for GEANT staff members during the outage because they all use GEANT Staff IdP for authentication:
- SharePoint (e.g. Intranet, Partner Portal)
- GEANT wiki
- EventR
- wordpress sites
- Compendium
- Filesender
- FoD
- Lifesize
- BOX
- sympa (lists at prod-lists01.geant.net server)
Cause
- Failure to follow the change management process - needs further investigation.
- Lack of planning. Preparing an isolated environment would have taken up to a week of work but we have been asked to make this work the same day. Massimiliano Adamo and Michael Haller asked to postpone and to meet on Friday.
- Massimiliano Adamo asked twice in the meeting for objections and if we intended to take an action. It was a team decision.
- severe lack of knowledge of puppet integration with CloudBolt, from the side of the consultants. The consultants have been asked repeatedly certain questions and they were either evading the answer or providing wrong replies (i.e.: CloudBolt doesn't provide Agent Bootstrap, but they said it does This implies that they don't know the implementation details). The consultants have been asked clearly about the environment settings.
agents must be pre-installed and configured in the image (from CloudBolt documentation: http://docs.cloudbolt.io/configuration-managers/puppet/index.html ) invetigation.
Timeline
Time (CET) | ||
---|---|---|
03 Aug, 12:36 | Issue Reported by Cristian Bandea on slack channel #techies | |
03 Aug, 13:02 | Andrew Jarvis sent direct Slack to Dick Visser and Konstantin Lepikhov but no response. | |
03 Aug, 13:25 | Massimiliano Adamo, started investigating | |
03 Aug, 13:35 | Andrew Jarvis contacted Massimiliano Adamo | , whom then investigated.|
03 Aug, 13:46 | user-7da5d pointed out to CloudBolt and found the issue (wrong puppet environment) | |
03 Aug, 14:00 | user-7da5d have switched off prod-idp01 VM leaving only prod-idp02 functioning | |
03 Aug, 16:00 | Saltstack was used to fix all the servers at once | |
...